I’m a data scientist who cares most about simplicity and optimisation: finding the cleanest, most efficient path from a messy problem to an answer people can trust.
Day to day, I shape ideas into features for platforms that handle high-volume financial data, working in R, Python and SQL, from forecasting to machine learning. I also design the prompts that guide AI agents through pipeline work. Outside work, I learn by building: my latest project is InkDoc, an open-source app that converts documents into clean Markdown for LLM and RAG pipelines. One rule runs through all of it: if it can’t be reconciled, it isn’t done.
All open source — code on GitHub.
A desktop app and local API that turns any document into clean, AI-ready Markdown, with most conversions running on your own machine.
Compared naïve, seasonal naïve, ETS, ARIMA and neural-network models on IBM’s monthly adjusted close, trained on 2016–May 2021 and tested on the following 7 months.
ARIMA cut test RMSE from 17.7 to 5.5 (MASE 0.37)
Worked solutions to Hyndman & Athanasopoulos: regression, time-series decomposition, exponential smoothing and ARIMA.
A secure-by-default Electron starter that wraps web apps in a frameless desktop window with a custom title bar.
Why splitting ingestion, quality and governance across separate platforms doesn't solve pipeline friction — stress-tested against the best objections to centralizing it, and the two honest conditions where it actually matters.
Why self-driving cars could save thousands of lives, nine ways they could be watching you, and the big bets, bold claims.